plugin [nlsearch] was built for Elasticsearch version 9.5.5 but version 9.5.4 is running
Install the zip for your version, or build one: ./gradlew bundle -PesVersion=9.5.4.
502, could not get an answer from ollama/qwen2.5-coder:7b: ... Connection refused
Ollama is not running, or runs somewhere else. Start it (ollama serve) or set
nlsearch.url.
502, ... model "qwen2.5-coder:7b" not found
ollama pull qwen2.5-coder:7b, or set nlsearch.model to one you have
(ollama list).
502, the model's answer is not valid JSON
Happens with very small models. Try qwen2.5-coder:7b or bigger, or another
provider. The raw answer is in the error so you can see what it did.
400, the model's plan for [update] has no "id"
The model wanted to update by id but had none. Rephrase (“set the price of
product 1 to 20”) or let it use a query (“set the price of the blue mug to 20”).
It picked the wrong index Name the index in the prompt (“in products, …”). With more than 20 indices the prompt only carries names, so field based guessing stops working; say which index you mean.
Slow answers
The first call after a restart loads the model (10 to 30 seconds with Ollama).
After that a 7B model answers in 1 to 4 seconds on an Apple Silicon laptop.
Big mappings make it slower; nlsearch.timeout is 60s by default.
SLF4J(W): No SLF4J providers were found in the Elasticsearch log
Three lines, once, the first time the plugin talks to a model. langchain4j
logs through SLF4J and there is no logging backend inside the plugin. Harmless.
Nothing in the logs The plugin only logs when it could not save history or send an error. Look at the response, everything is in there: the model used, the plan, the result.